Thesaurus Support when Searching Earth Science Data
James C. French Lola M. Olsen Worthy N. Martin Department of Computer Science Code 902, Bldg.32, S130D Department of Computer Science University of Virginia NASA/GSFC University of Virginia Charlottesville, VA Greenbelt, MD 20771 Charlottesville, VA
[email protected] [email protected] [email protected] Abstract | Keyword hierarchies are known to assist in searches for Earth science data sets. The level of assistance, however, is dependent on the common semantic interpretation by the indexer and the searcher. A strategy to improve search results may lie in the use of the elusive thesaurus. Thesauri have been discussed as the solution to semantic interpretation over the years. However, integrating their use within an interactive search has proven to be more dicult, as the thesauri built to date have most often been stand-alone. To negotiate this hurdle, we describe the integration of a thesaurus built on a well-functioning operational locator of Earth science data.
T
I. Introduction
HE EOSDIS is a multidisciplinary data store. This leads to diculties in cross-disciplinary searching due mainly to di erences in terminology. This may manifest itself in several di erent ways. Users may be faced with unfamiliar jargon when searching in another discipline. They may also use a term that has a di erent meaning in another discipline. Consider, for example, the term \aerosol." It might refer to gases only or particulate matter or both. There is no way a priori to know what a user means by the term, what is included, or what is excluded. The speci c semantic diculty is that the system indexes using one vocabulary and it might be quite di erent from the vocabulary being employed by any particular searcher. One attack on this problem is to provide a controlled vocabulary for use in searching. This approach has been used very e ectively in some Earth science data systems, for example, the GCMD1 (Global Change Master Directory). Another approach to mitigate this problem is to provide a thesaurus to help suggest useful search terms to searchers. The work reported here is focused on the latter approach. The full context of our This work supported in part by NASA Grants NAG5-8585 and NAG5-9747 and NASA GSRP NGT5-50062. 1 http://gcmd.nasa.gov/
research on these problems is outlined in [2]. In a companion paper [1] we describe our conceptual framework and additional approaches to mitigate these vocabulary problems. Our current prototype work has developed and demonstrated an integrated thesaurus service for Earth science data systems. Our initial prototype was demonstrated in connection with the GCMD. One of our objectives is to provide an integrated thesaurus server that can be accessed from other NASA ES data systems such as the EOS Data Gateway (EDG). Term suggestion and thesaurus support can be useful in any interface. II. DLR Thesaurus
The German Remote Sensing Data Center (DFD)2 of the German Aerospace Center (DLR)3 developed the original thesaurus that forms the foundation of this work. The DLR thesaurus uses Oracle 8i on the server-side to manage the thesaurus data structure. A fragment of that data structure is depicted in Figure 1. This data structure captures the important thesaural notions by conceptually linking terms in a graph with appropriate relationships, e.g., synonyms, broader and narrower terms, and related terms. Each node in Figure 1 is shown with a label and the number of synonyms contained in the node. In a sense the node is known by n + 1 equivalent labels. Note that the DLR thesaurus contains English and German synonyms but our counts only show the count of English synonyms. This is to give the reader an idea of the richness of the thesaurus in a monolingual mode. A. Search Assistant We have written a new client-side Java applet to access the thesaurus. A screen shot of our interface is 2 http://www.dfd.dlr.de 3 http://www.dlr.de
shown in Figure 2. The screen shown in the gure is in response to a user query for the string \atmospheric pollution." Note that the entry is labeled \air-pollution." Any of the synonyms listed will retrieve this node; the string \air-pollution" has simply been designated as the node's canonical name. The display of Figure 2 is \located at" the air pollution node in Figure 1. The bold lines shown in the data structure (Figure 1) correspond to the terms enumerated in the display (Figure 2). The thesaurus data structure provides for the following relationships. They are described with respect to the current concept. For examples of each refer to Figure 2.
or updating existing terms. The Update Assistant is intended for restricted access by personnel responsible for the thesaurus maintenance. III. Deployment Strategy
In our initial prototype we provide for direct access to the thesaurus service as shown in Figure 4. The thesaurus service is stateless with respect to the invoking interface. The client-side Search Assistant is responsible for managing the modi ed query string. The speci c ES data system is unaware of the existence of the thesaurus server. We are currently reworking the thesaurus interface so that it can be packaged as a web service and exported to ES data system applications via ECHO (EOSDIS Synonyms: An equivalence class of strings denoting this ClearingHOuse)[3]. concept. One of the strings is used as a label for the References class. [1] J. C. French, A. C. Chapin, and W. N. Martin. Using Multiple Top terms: The terms at the top of the hierarchy. Viewpoints to Improve Access to Earth Science Data. In Proc. Earth Science Technology Conference, 2002. These are the broadest terms containing this concept. [2] J. C. French, W. N. Martin, and L. M. Olsen. Extending Broader: Immediate predecessor terms in the hierarchy. the Vocabulary Available for Cross-Disciplinary Searching of Earth Science Data. Technical Report CS-2002-04, DepartNarrower: Immediate successor terms in the hierarchy. of Computer Science, University of Virginia, 2002. Related: Arbitrary terms in the thesaurus structure. [3] ment R. P ster, R. Ullman, and K. Wichmann. ECHO Responds Provide an alternative method for navigating the structo NASA's Earth Science User Community. In HCI International, 2001. ture. Each of these categories is represented in the display for a concept, for example see Figure 2. Our interface supports two main activities: 1. Adding terms to the current query. Any term in any category can be added to the current query. Simply mouse over the term and right-click the mouse. The selected term is added to the query. 2. Navigating the thesaurus structure. The display can be refocused to any of the latter four categories from above (top terms, broader, narrower, or related). Simply mouse over the term and left-click the mouse. When the Finish button is clicked, the Search Assistant returns to the form from which it was invoked with the modi ed search string substituted for the initial search string. Figure 3 shows how simply the thesaurus Search Assistant can be implemented into the GCMD interface. B. Update Assistant We have also prototyped an Update Assistant to facilitate maintenance of the thesaurus data structure. The interface, not shown here, is very similar to the Search Assistant and provides familiar add, change, delete functionality for introducing new thesaurus terms
...
global change (4)
...
pollution (6)
...
...
...
acidification (1)
food contamination (2)
air pollution (22)
global warming (5)
...
...
indoor pollution (1)
... trace gases (11)
...
...
aerosols (5)
carbon monoxide (6)
...
...
...
NOx (1)
...
sulfur dioxide (6)
Fig. 1. DLR thesaurus data structure.
Fig. 2. Interface to Search Assistant.
...
... air quality (2)
Fig. 3. GCMD interface with thesaurus Search Assistant added.
Thesaurus Service
EOS Data Gateway
GCMD
DATA DATA
DATA
DATA
...
DATA
DAAC
DAAC
...
DAAC
Fig. 4. Stand-alone server accessed directly from applications.
service service
Thesaurus Service
Thesaurus Interface
ECHO
DAAC Interface
service
EOS Data Gateway
GCMD
DATA DATA
DATA
DATA
...
DATA
Fig. 5. Stand-alone server with direct access from applications and also packaged as a web service via ECHO.
